Vectorizer.raspberry

🚀 Model Card for vectorizer.raspberry

This model, developed by Sinequa, is a vectorizer that generates an embedding vector for a given passage or query. Passage vectors are stored in our vector index, and the query vector is used at query time to search for relevant passages in the index.

🚀 Quick Start

This model is ready to use for generating embedding vectors for passages and queries. You can start leveraging its capabilities right away.

✨ Features

• Multilingual Support: Trained and tested in multiple languages including English, French, German, Spanish, Italian, Dutch, Japanese, Portuguese, and Simplified Chinese. Also offers basic support for 91 additional languages used in the base model's pretraining.
• Efficient Inference: Provides fast inference times on various NVIDIA GPUs with different quantization types and batch sizes.
• Low Memory Consumption: Consumes relatively low GPU memory, with clear details provided for different quantization types.

📦 Installation

Requirements

• Minimal Sinequa version: 11.10.0
• Minimal Sinequa version for using FP16 models and GPUs with CUDA compute capability of 8.9+ (like NVIDIA L4): 11.11.0
• Cuda compute capability: above 5.0 (above 6.0 for FP16 use)

💻 Usage Examples

Basic Usage

The model can be used to generate embedding vectors for passages and queries. Here's a high - level concept of how it might be used in a Python - like environment (actual implementation may vary based on the integration):

# Assume there is a function provided by Sinequa to interact with the model
from sinequa_vectorizer import Vectorizer

vectorizer = Vectorizer(model_name='vectorizer.raspberry')
passage = "This is a sample passage."
passage_vector = vectorizer.get_vector(passage)

📚 Documentation

Supported Languages

The model was trained and tested in the following languages:

• English
• French
• German
• Spanish
• Italian
• Dutch
• Japanese
• Portuguese
• Chinese (simplified)

Besides these languages, basic support can be expected for additional 91 languages that were used during the pretraining of the base model (see Appendix A of XLM - R paper).

Scores

Note that the relevance score is computed as an average over 14 retrieval datasets (see details below).

Inference Times

GPU Memory usage

Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which can be around 0.5 to 1 GiB depending on the used GPU.

Model Details

Overview

Training Data

The model has been trained using all datasets that are cited in the [all - MiniLM - L6 - v2](https://huggingface.co/sentence - transformers/all - MiniLM - L6 - v2) model. In addition to that, this model has been trained on the datasets cited in this paper on the 9 aforementioned languages.

Evaluation Metrics

To determine the relevance score, we averaged the results that we obtained when evaluating on the datasets of the [BEIR benchmark](https://github.com/beir - cellar/beir). Note that all these datasets are in English.

We evaluated the model on the datasets of the [MIRACL benchmark](https://github.com/project - miracl/miracl) to test its multilingual capacities. Note that not all training languages are part of the benchmark, so we only report the metrics for the existing languages.

🔧 Technical Details

The model is based on the [mMiniLMv2 - L6 - H384 - distilled - from - XLMR - Large](https://huggingface.co/nreimers/mMiniLMv2 - L6 - H384 - distilled - from - XLMR - Large) base model. It has 107 million parameters and is designed to be insensitive to casing and accents. The output dimensions are reduced to 256 using an additional dense layer. The training procedure involves different strategies for datasets with and without mined hard negative data, and the number of negatives is augmented using an in - batch negative strategy.